Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Debug helper to dump the current kernel pagetables of the system
4 * so that we can see what the various memory ranges are set to.
5 *
6 * (C) Copyright 2008 Intel Corporation
7 *
8 * Author: Arjan van de Ven <arjan@linux.intel.com>
9 */
10
11#include <linux/debugfs.h>
12#include <linux/kasan.h>
13#include <linux/mm.h>
14#include <linux/init.h>
15#include <linux/sched.h>
16#include <linux/seq_file.h>
17#include <linux/highmem.h>
18#include <linux/pci.h>
19
20#include <asm/e820/types.h>
21#include <asm/pgtable.h>
22
23/*
24 * The dumper groups pagetable entries of the same type into one, and for
25 * that it needs to keep some state when walking, and flush this state
26 * when a "break" in the continuity is found.
27 */
28struct pg_state {
29 int level;
30 pgprot_t current_prot;
31 pgprotval_t effective_prot;
32 unsigned long start_address;
33 unsigned long current_address;
34 const struct addr_marker *marker;
35 unsigned long lines;
36 bool to_dmesg;
37 bool check_wx;
38 unsigned long wx_pages;
39};
40
41struct addr_marker {
42 unsigned long start_address;
43 const char *name;
44 unsigned long max_lines;
45};
46
47/* Address space markers hints */
48
49#ifdef CONFIG_X86_64
50
51enum address_markers_idx {
52 USER_SPACE_NR = 0,
53 KERNEL_SPACE_NR,
54#ifdef CONFIG_MODIFY_LDT_SYSCALL
55 LDT_NR,
56#endif
57 LOW_KERNEL_NR,
58 VMALLOC_START_NR,
59 VMEMMAP_START_NR,
60#ifdef CONFIG_KASAN
61 KASAN_SHADOW_START_NR,
62 KASAN_SHADOW_END_NR,
63#endif
64 CPU_ENTRY_AREA_NR,
65#ifdef CONFIG_X86_ESPFIX64
66 ESPFIX_START_NR,
67#endif
68#ifdef CONFIG_EFI
69 EFI_END_NR,
70#endif
71 HIGH_KERNEL_NR,
72 MODULES_VADDR_NR,
73 MODULES_END_NR,
74 FIXADDR_START_NR,
75 END_OF_SPACE_NR,
76};
77
78static struct addr_marker address_markers[] = {
79 [USER_SPACE_NR] = { 0, "User Space" },
80 [KERNEL_SPACE_NR] = { (1UL << 63), "Kernel Space" },
81 [LOW_KERNEL_NR] = { 0UL, "Low Kernel Mapping" },
82 [VMALLOC_START_NR] = { 0UL, "vmalloc() Area" },
83 [VMEMMAP_START_NR] = { 0UL, "Vmemmap" },
84#ifdef CONFIG_KASAN
85 /*
86 * These fields get initialized with the (dynamic)
87 * KASAN_SHADOW_{START,END} values in pt_dump_init().
88 */
89 [KASAN_SHADOW_START_NR] = { 0UL, "KASAN shadow" },
90 [KASAN_SHADOW_END_NR] = { 0UL, "KASAN shadow end" },
91#endif
92#ifdef CONFIG_MODIFY_LDT_SYSCALL
93 [LDT_NR] = { 0UL, "LDT remap" },
94#endif
95 [CPU_ENTRY_AREA_NR] = { CPU_ENTRY_AREA_BASE,"CPU entry Area" },
96#ifdef CONFIG_X86_ESPFIX64
97 [ESPFIX_START_NR] = { ESPFIX_BASE_ADDR, "ESPfix Area", 16 },
98#endif
99#ifdef CONFIG_EFI
100 [EFI_END_NR] = { EFI_VA_END, "EFI Runtime Services" },
101#endif
102 [HIGH_KERNEL_NR] = { __START_KERNEL_map, "High Kernel Mapping" },
103 [MODULES_VADDR_NR] = { MODULES_VADDR, "Modules" },
104 [MODULES_END_NR] = { MODULES_END, "End Modules" },
105 [FIXADDR_START_NR] = { FIXADDR_START, "Fixmap Area" },
106 [END_OF_SPACE_NR] = { -1, NULL }
107};
108
109#define INIT_PGD ((pgd_t *) &init_top_pgt)
110
111#else /* CONFIG_X86_64 */
112
113enum address_markers_idx {
114 USER_SPACE_NR = 0,
115 KERNEL_SPACE_NR,
116 VMALLOC_START_NR,
117 VMALLOC_END_NR,
118#ifdef CONFIG_HIGHMEM
119 PKMAP_BASE_NR,
120#endif
121#ifdef CONFIG_MODIFY_LDT_SYSCALL
122 LDT_NR,
123#endif
124 CPU_ENTRY_AREA_NR,
125 FIXADDR_START_NR,
126 END_OF_SPACE_NR,
127};
128
129static struct addr_marker address_markers[] = {
130 [USER_SPACE_NR] = { 0, "User Space" },
131 [KERNEL_SPACE_NR] = { PAGE_OFFSET, "Kernel Mapping" },
132 [VMALLOC_START_NR] = { 0UL, "vmalloc() Area" },
133 [VMALLOC_END_NR] = { 0UL, "vmalloc() End" },
134#ifdef CONFIG_HIGHMEM
135 [PKMAP_BASE_NR] = { 0UL, "Persistent kmap() Area" },
136#endif
137#ifdef CONFIG_MODIFY_LDT_SYSCALL
138 [LDT_NR] = { 0UL, "LDT remap" },
139#endif
140 [CPU_ENTRY_AREA_NR] = { 0UL, "CPU entry area" },
141 [FIXADDR_START_NR] = { 0UL, "Fixmap area" },
142 [END_OF_SPACE_NR] = { -1, NULL }
143};
144
145#define INIT_PGD (swapper_pg_dir)
146
147#endif /* !CONFIG_X86_64 */
148
149/* Multipliers for offsets within the PTEs */
150#define PTE_LEVEL_MULT (PAGE_SIZE)
151#define PMD_LEVEL_MULT (PTRS_PER_PTE * PTE_LEVEL_MULT)
152#define PUD_LEVEL_MULT (PTRS_PER_PMD * PMD_LEVEL_MULT)
153#define P4D_LEVEL_MULT (PTRS_PER_PUD * PUD_LEVEL_MULT)
154#define PGD_LEVEL_MULT (PTRS_PER_P4D * P4D_LEVEL_MULT)
155
156#define pt_dump_seq_printf(m, to_dmesg, fmt, args...) \
157({ \
158 if (to_dmesg) \
159 printk(KERN_INFO fmt, ##args); \
160 else \
161 if (m) \
162 seq_printf(m, fmt, ##args); \
163})
164
165#define pt_dump_cont_printf(m, to_dmesg, fmt, args...) \
166({ \
167 if (to_dmesg) \
168 printk(KERN_CONT fmt, ##args); \
169 else \
170 if (m) \
171 seq_printf(m, fmt, ##args); \
172})
173
174/*
175 * Print a readable form of a pgprot_t to the seq_file
176 */
177static void printk_prot(struct seq_file *m, pgprot_t prot, int level, bool dmsg)
178{
179 pgprotval_t pr = pgprot_val(prot);
180 static const char * const level_name[] =
181 { "cr3", "pgd", "p4d", "pud", "pmd", "pte" };
182
183 if (!(pr & _PAGE_PRESENT)) {
184 /* Not present */
185 pt_dump_cont_printf(m, dmsg, " ");
186 } else {
187 if (pr & _PAGE_USER)
188 pt_dump_cont_printf(m, dmsg, "USR ");
189 else
190 pt_dump_cont_printf(m, dmsg, " ");
191 if (pr & _PAGE_RW)
192 pt_dump_cont_printf(m, dmsg, "RW ");
193 else
194 pt_dump_cont_printf(m, dmsg, "ro ");
195 if (pr & _PAGE_PWT)
196 pt_dump_cont_printf(m, dmsg, "PWT ");
197 else
198 pt_dump_cont_printf(m, dmsg, " ");
199 if (pr & _PAGE_PCD)
200 pt_dump_cont_printf(m, dmsg, "PCD ");
201 else
202 pt_dump_cont_printf(m, dmsg, " ");
203
204 /* Bit 7 has a different meaning on level 3 vs 4 */
205 if (level <= 4 && pr & _PAGE_PSE)
206 pt_dump_cont_printf(m, dmsg, "PSE ");
207 else
208 pt_dump_cont_printf(m, dmsg, " ");
209 if ((level == 5 && pr & _PAGE_PAT) ||
210 ((level == 4 || level == 3) && pr & _PAGE_PAT_LARGE))
211 pt_dump_cont_printf(m, dmsg, "PAT ");
212 else
213 pt_dump_cont_printf(m, dmsg, " ");
214 if (pr & _PAGE_GLOBAL)
215 pt_dump_cont_printf(m, dmsg, "GLB ");
216 else
217 pt_dump_cont_printf(m, dmsg, " ");
218 if (pr & _PAGE_NX)
219 pt_dump_cont_printf(m, dmsg, "NX ");
220 else
221 pt_dump_cont_printf(m, dmsg, "x ");
222 }
223 pt_dump_cont_printf(m, dmsg, "%s\n", level_name[level]);
224}
225
226/*
227 * On 64 bits, sign-extend the 48 bit address to 64 bit
228 */
229static unsigned long normalize_addr(unsigned long u)
230{
231 int shift;
232 if (!IS_ENABLED(CONFIG_X86_64))
233 return u;
234
235 shift = 64 - (__VIRTUAL_MASK_SHIFT + 1);
236 return (signed long)(u << shift) >> shift;
237}
238
239static void note_wx(struct pg_state *st)
240{
241 unsigned long npages;
242
243 npages = (st->current_address - st->start_address) / PAGE_SIZE;
244
245#ifdef CONFIG_PCI_BIOS
246 /*
247 * If PCI BIOS is enabled, the PCI BIOS area is forced to WX.
248 * Inform about it, but avoid the warning.
249 */
250 if (pcibios_enabled && st->start_address >= PAGE_OFFSET + BIOS_BEGIN &&
251 st->current_address <= PAGE_OFFSET + BIOS_END) {
252 pr_warn_once("x86/mm: PCI BIOS W+X mapping %lu pages\n", npages);
253 return;
254 }
255#endif
256 /* Account the WX pages */
257 st->wx_pages += npages;
258 WARN_ONCE(__supported_pte_mask & _PAGE_NX,
259 "x86/mm: Found insecure W+X mapping at address %pS\n",
260 (void *)st->start_address);
261}
262
263/*
264 * This function gets called on a break in a continuous series
265 * of PTE entries; the next one is different so we need to
266 * print what we collected so far.
267 */
268static void note_page(struct seq_file *m, struct pg_state *st,
269 pgprot_t new_prot, pgprotval_t new_eff, int level)
270{
271 pgprotval_t prot, cur, eff;
272 static const char units[] = "BKMGTPE";
273
274 /*
275 * If we have a "break" in the series, we need to flush the state that
276 * we have now. "break" is either changing perms, levels or
277 * address space marker.
278 */
279 prot = pgprot_val(new_prot);
280 cur = pgprot_val(st->current_prot);
281 eff = st->effective_prot;
282
283 if (!st->level) {
284 /* First entry */
285 st->current_prot = new_prot;
286 st->effective_prot = new_eff;
287 st->level = level;
288 st->marker = address_markers;
289 st->lines = 0;
290 pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
291 st->marker->name);
292 } else if (prot != cur || new_eff != eff || level != st->level ||
293 st->current_address >= st->marker[1].start_address) {
294 const char *unit = units;
295 unsigned long delta;
296 int width = sizeof(unsigned long) * 2;
297
298 if (st->check_wx && (eff & _PAGE_RW) && !(eff & _PAGE_NX))
299 note_wx(st);
300
301 /*
302 * Now print the actual finished series
303 */
304 if (!st->marker->max_lines ||
305 st->lines < st->marker->max_lines) {
306 pt_dump_seq_printf(m, st->to_dmesg,
307 "0x%0*lx-0x%0*lx ",
308 width, st->start_address,
309 width, st->current_address);
310
311 delta = st->current_address - st->start_address;
312 while (!(delta & 1023) && unit[1]) {
313 delta >>= 10;
314 unit++;
315 }
316 pt_dump_cont_printf(m, st->to_dmesg, "%9lu%c ",
317 delta, *unit);
318 printk_prot(m, st->current_prot, st->level,
319 st->to_dmesg);
320 }
321 st->lines++;
322
323 /*
324 * We print markers for special areas of address space,
325 * such as the start of vmalloc space etc.
326 * This helps in the interpretation.
327 */
328 if (st->current_address >= st->marker[1].start_address) {
329 if (st->marker->max_lines &&
330 st->lines > st->marker->max_lines) {
331 unsigned long nskip =
332 st->lines - st->marker->max_lines;
333 pt_dump_seq_printf(m, st->to_dmesg,
334 "... %lu entr%s skipped ... \n",
335 nskip,
336 nskip == 1 ? "y" : "ies");
337 }
338 st->marker++;
339 st->lines = 0;
340 pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
341 st->marker->name);
342 }
343
344 st->start_address = st->current_address;
345 st->current_prot = new_prot;
346 st->effective_prot = new_eff;
347 st->level = level;
348 }
349}
350
351static inline pgprotval_t effective_prot(pgprotval_t prot1, pgprotval_t prot2)
352{
353 return (prot1 & prot2 & (_PAGE_USER | _PAGE_RW)) |
354 ((prot1 | prot2) & _PAGE_NX);
355}
356
357static void walk_pte_level(struct seq_file *m, struct pg_state *st, pmd_t addr,
358 pgprotval_t eff_in, unsigned long P)
359{
360 int i;
361 pte_t *pte;
362 pgprotval_t prot, eff;
363
364 for (i = 0; i < PTRS_PER_PTE; i++) {
365 st->current_address = normalize_addr(P + i * PTE_LEVEL_MULT);
366 pte = pte_offset_map(&addr, st->current_address);
367 prot = pte_flags(*pte);
368 eff = effective_prot(eff_in, prot);
369 note_page(m, st, __pgprot(prot), eff, 5);
370 pte_unmap(pte);
371 }
372}
373#ifdef CONFIG_KASAN
374
375/*
376 * This is an optimization for KASAN=y case. Since all kasan page tables
377 * eventually point to the kasan_early_shadow_page we could call note_page()
378 * right away without walking through lower level page tables. This saves
379 * us dozens of seconds (minutes for 5-level config) while checking for
380 * W+X mapping or reading kernel_page_tables debugfs file.
381 */
382static inline bool kasan_page_table(struct seq_file *m, struct pg_state *st,
383 void *pt)
384{
385 if (__pa(pt) == __pa(kasan_early_shadow_pmd) ||
386 (pgtable_l5_enabled() &&
387 __pa(pt) == __pa(kasan_early_shadow_p4d)) ||
388 __pa(pt) == __pa(kasan_early_shadow_pud)) {
389 pgprotval_t prot = pte_flags(kasan_early_shadow_pte[0]);
390 note_page(m, st, __pgprot(prot), 0, 5);
391 return true;
392 }
393 return false;
394}
395#else
396static inline bool kasan_page_table(struct seq_file *m, struct pg_state *st,
397 void *pt)
398{
399 return false;
400}
401#endif
402
403#if PTRS_PER_PMD > 1
404
405static void walk_pmd_level(struct seq_file *m, struct pg_state *st, pud_t addr,
406 pgprotval_t eff_in, unsigned long P)
407{
408 int i;
409 pmd_t *start, *pmd_start;
410 pgprotval_t prot, eff;
411
412 pmd_start = start = (pmd_t *)pud_page_vaddr(addr);
413 for (i = 0; i < PTRS_PER_PMD; i++) {
414 st->current_address = normalize_addr(P + i * PMD_LEVEL_MULT);
415 if (!pmd_none(*start)) {
416 prot = pmd_flags(*start);
417 eff = effective_prot(eff_in, prot);
418 if (pmd_large(*start) || !pmd_present(*start)) {
419 note_page(m, st, __pgprot(prot), eff, 4);
420 } else if (!kasan_page_table(m, st, pmd_start)) {
421 walk_pte_level(m, st, *start, eff,
422 P + i * PMD_LEVEL_MULT);
423 }
424 } else
425 note_page(m, st, __pgprot(0), 0, 4);
426 start++;
427 }
428}
429
430#else
431#define walk_pmd_level(m,s,a,e,p) walk_pte_level(m,s,__pmd(pud_val(a)),e,p)
432#define pud_large(a) pmd_large(__pmd(pud_val(a)))
433#define pud_none(a) pmd_none(__pmd(pud_val(a)))
434#endif
435
436#if PTRS_PER_PUD > 1
437
438static void walk_pud_level(struct seq_file *m, struct pg_state *st, p4d_t addr,
439 pgprotval_t eff_in, unsigned long P)
440{
441 int i;
442 pud_t *start, *pud_start;
443 pgprotval_t prot, eff;
444
445 pud_start = start = (pud_t *)p4d_page_vaddr(addr);
446
447 for (i = 0; i < PTRS_PER_PUD; i++) {
448 st->current_address = normalize_addr(P + i * PUD_LEVEL_MULT);
449 if (!pud_none(*start)) {
450 prot = pud_flags(*start);
451 eff = effective_prot(eff_in, prot);
452 if (pud_large(*start) || !pud_present(*start)) {
453 note_page(m, st, __pgprot(prot), eff, 3);
454 } else if (!kasan_page_table(m, st, pud_start)) {
455 walk_pmd_level(m, st, *start, eff,
456 P + i * PUD_LEVEL_MULT);
457 }
458 } else
459 note_page(m, st, __pgprot(0), 0, 3);
460
461 start++;
462 }
463}
464
465#else
466#define walk_pud_level(m,s,a,e,p) walk_pmd_level(m,s,__pud(p4d_val(a)),e,p)
467#define p4d_large(a) pud_large(__pud(p4d_val(a)))
468#define p4d_none(a) pud_none(__pud(p4d_val(a)))
469#endif
470
471static void walk_p4d_level(struct seq_file *m, struct pg_state *st, pgd_t addr,
472 pgprotval_t eff_in, unsigned long P)
473{
474 int i;
475 p4d_t *start, *p4d_start;
476 pgprotval_t prot, eff;
477
478 if (PTRS_PER_P4D == 1)
479 return walk_pud_level(m, st, __p4d(pgd_val(addr)), eff_in, P);
480
481 p4d_start = start = (p4d_t *)pgd_page_vaddr(addr);
482
483 for (i = 0; i < PTRS_PER_P4D; i++) {
484 st->current_address = normalize_addr(P + i * P4D_LEVEL_MULT);
485 if (!p4d_none(*start)) {
486 prot = p4d_flags(*start);
487 eff = effective_prot(eff_in, prot);
488 if (p4d_large(*start) || !p4d_present(*start)) {
489 note_page(m, st, __pgprot(prot), eff, 2);
490 } else if (!kasan_page_table(m, st, p4d_start)) {
491 walk_pud_level(m, st, *start, eff,
492 P + i * P4D_LEVEL_MULT);
493 }
494 } else
495 note_page(m, st, __pgprot(0), 0, 2);
496
497 start++;
498 }
499}
500
501#define pgd_large(a) (pgtable_l5_enabled() ? pgd_large(a) : p4d_large(__p4d(pgd_val(a))))
502#define pgd_none(a) (pgtable_l5_enabled() ? pgd_none(a) : p4d_none(__p4d(pgd_val(a))))
503
504static inline bool is_hypervisor_range(int idx)
505{
506#ifdef CONFIG_X86_64
507 /*
508 * A hole in the beginning of kernel address space reserved
509 * for a hypervisor.
510 */
511 return (idx >= pgd_index(GUARD_HOLE_BASE_ADDR)) &&
512 (idx < pgd_index(GUARD_HOLE_END_ADDR));
513#else
514 return false;
515#endif
516}
517
518static void ptdump_walk_pgd_level_core(struct seq_file *m, pgd_t *pgd,
519 bool checkwx, bool dmesg)
520{
521 pgd_t *start = INIT_PGD;
522 pgprotval_t prot, eff;
523 int i;
524 struct pg_state st = {};
525
526 if (pgd) {
527 start = pgd;
528 st.to_dmesg = dmesg;
529 }
530
531 st.check_wx = checkwx;
532 if (checkwx)
533 st.wx_pages = 0;
534
535 for (i = 0; i < PTRS_PER_PGD; i++) {
536 st.current_address = normalize_addr(i * PGD_LEVEL_MULT);
537 if (!pgd_none(*start) && !is_hypervisor_range(i)) {
538 prot = pgd_flags(*start);
539#ifdef CONFIG_X86_PAE
540 eff = _PAGE_USER | _PAGE_RW;
541#else
542 eff = prot;
543#endif
544 if (pgd_large(*start) || !pgd_present(*start)) {
545 note_page(m, &st, __pgprot(prot), eff, 1);
546 } else {
547 walk_p4d_level(m, &st, *start, eff,
548 i * PGD_LEVEL_MULT);
549 }
550 } else
551 note_page(m, &st, __pgprot(0), 0, 1);
552
553 cond_resched();
554 start++;
555 }
556
557 /* Flush out the last page */
558 st.current_address = normalize_addr(PTRS_PER_PGD*PGD_LEVEL_MULT);
559 note_page(m, &st, __pgprot(0), 0, 0);
560 if (!checkwx)
561 return;
562 if (st.wx_pages)
563 pr_info("x86/mm: Checked W+X mappings: FAILED, %lu W+X pages found.\n",
564 st.wx_pages);
565 else
566 pr_info("x86/mm: Checked W+X mappings: passed, no W+X pages found.\n");
567}
568
569void ptdump_walk_pgd_level(struct seq_file *m, pgd_t *pgd)
570{
571 ptdump_walk_pgd_level_core(m, pgd, false, true);
572}
573
574void ptdump_walk_pgd_level_debugfs(struct seq_file *m, pgd_t *pgd, bool user)
575{
576#ifdef CONFIG_PAGE_TABLE_ISOLATION
577 if (user && boot_cpu_has(X86_FEATURE_PTI))
578 pgd = kernel_to_user_pgdp(pgd);
579#endif
580 ptdump_walk_pgd_level_core(m, pgd, false, false);
581}
582EXPORT_SYMBOL_GPL(ptdump_walk_pgd_level_debugfs);
583
584void ptdump_walk_user_pgd_level_checkwx(void)
585{
586#ifdef CONFIG_PAGE_TABLE_ISOLATION
587 pgd_t *pgd = INIT_PGD;
588
589 if (!(__supported_pte_mask & _PAGE_NX) ||
590 !boot_cpu_has(X86_FEATURE_PTI))
591 return;
592
593 pr_info("x86/mm: Checking user space page tables\n");
594 pgd = kernel_to_user_pgdp(pgd);
595 ptdump_walk_pgd_level_core(NULL, pgd, true, false);
596#endif
597}
598
599void ptdump_walk_pgd_level_checkwx(void)
600{
601 ptdump_walk_pgd_level_core(NULL, NULL, true, false);
602}
603
604static int __init pt_dump_init(void)
605{
606 /*
607 * Various markers are not compile-time constants, so assign them
608 * here.
609 */
610#ifdef CONFIG_X86_64
611 address_markers[LOW_KERNEL_NR].start_address = PAGE_OFFSET;
612 address_markers[VMALLOC_START_NR].start_address = VMALLOC_START;
613 address_markers[VMEMMAP_START_NR].start_address = VMEMMAP_START;
614#ifdef CONFIG_MODIFY_LDT_SYSCALL
615 address_markers[LDT_NR].start_address = LDT_BASE_ADDR;
616#endif
617#ifdef CONFIG_KASAN
618 address_markers[KASAN_SHADOW_START_NR].start_address = KASAN_SHADOW_START;
619 address_markers[KASAN_SHADOW_END_NR].start_address = KASAN_SHADOW_END;
620#endif
621#endif
622#ifdef CONFIG_X86_32
623 address_markers[VMALLOC_START_NR].start_address = VMALLOC_START;
624 address_markers[VMALLOC_END_NR].start_address = VMALLOC_END;
625# ifdef CONFIG_HIGHMEM
626 address_markers[PKMAP_BASE_NR].start_address = PKMAP_BASE;
627# endif
628 address_markers[FIXADDR_START_NR].start_address = FIXADDR_START;
629 address_markers[CPU_ENTRY_AREA_NR].start_address = CPU_ENTRY_AREA_BASE;
630# ifdef CONFIG_MODIFY_LDT_SYSCALL
631 address_markers[LDT_NR].start_address = LDT_BASE_ADDR;
632# endif
633#endif
634 return 0;
635}
636__initcall(pt_dump_init);
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Debug helper to dump the current kernel pagetables of the system
4 * so that we can see what the various memory ranges are set to.
5 *
6 * (C) Copyright 2008 Intel Corporation
7 *
8 * Author: Arjan van de Ven <arjan@linux.intel.com>
9 */
10
11#include <linux/debugfs.h>
12#include <linux/kasan.h>
13#include <linux/mm.h>
14#include <linux/init.h>
15#include <linux/sched.h>
16#include <linux/seq_file.h>
17#include <linux/highmem.h>
18#include <linux/pci.h>
19#include <linux/ptdump.h>
20
21#include <asm/e820/types.h>
22
23/*
24 * The dumper groups pagetable entries of the same type into one, and for
25 * that it needs to keep some state when walking, and flush this state
26 * when a "break" in the continuity is found.
27 */
28struct pg_state {
29 struct ptdump_state ptdump;
30 int level;
31 pgprotval_t current_prot;
32 pgprotval_t effective_prot;
33 pgprotval_t prot_levels[5];
34 unsigned long start_address;
35 const struct addr_marker *marker;
36 unsigned long lines;
37 bool to_dmesg;
38 bool check_wx;
39 unsigned long wx_pages;
40 struct seq_file *seq;
41};
42
43struct addr_marker {
44 unsigned long start_address;
45 const char *name;
46 unsigned long max_lines;
47};
48
49/* Address space markers hints */
50
51#ifdef CONFIG_X86_64
52
53enum address_markers_idx {
54 USER_SPACE_NR = 0,
55 KERNEL_SPACE_NR,
56#ifdef CONFIG_MODIFY_LDT_SYSCALL
57 LDT_NR,
58#endif
59 LOW_KERNEL_NR,
60 VMALLOC_START_NR,
61 VMEMMAP_START_NR,
62#ifdef CONFIG_KASAN
63 KASAN_SHADOW_START_NR,
64 KASAN_SHADOW_END_NR,
65#endif
66 CPU_ENTRY_AREA_NR,
67#ifdef CONFIG_X86_ESPFIX64
68 ESPFIX_START_NR,
69#endif
70#ifdef CONFIG_EFI
71 EFI_END_NR,
72#endif
73 HIGH_KERNEL_NR,
74 MODULES_VADDR_NR,
75 MODULES_END_NR,
76 FIXADDR_START_NR,
77 END_OF_SPACE_NR,
78};
79
80static struct addr_marker address_markers[] = {
81 [USER_SPACE_NR] = { 0, "User Space" },
82 [KERNEL_SPACE_NR] = { (1UL << 63), "Kernel Space" },
83 [LOW_KERNEL_NR] = { 0UL, "Low Kernel Mapping" },
84 [VMALLOC_START_NR] = { 0UL, "vmalloc() Area" },
85 [VMEMMAP_START_NR] = { 0UL, "Vmemmap" },
86#ifdef CONFIG_KASAN
87 /*
88 * These fields get initialized with the (dynamic)
89 * KASAN_SHADOW_{START,END} values in pt_dump_init().
90 */
91 [KASAN_SHADOW_START_NR] = { 0UL, "KASAN shadow" },
92 [KASAN_SHADOW_END_NR] = { 0UL, "KASAN shadow end" },
93#endif
94#ifdef CONFIG_MODIFY_LDT_SYSCALL
95 [LDT_NR] = { 0UL, "LDT remap" },
96#endif
97 [CPU_ENTRY_AREA_NR] = { CPU_ENTRY_AREA_BASE,"CPU entry Area" },
98#ifdef CONFIG_X86_ESPFIX64
99 [ESPFIX_START_NR] = { ESPFIX_BASE_ADDR, "ESPfix Area", 16 },
100#endif
101#ifdef CONFIG_EFI
102 [EFI_END_NR] = { EFI_VA_END, "EFI Runtime Services" },
103#endif
104 [HIGH_KERNEL_NR] = { __START_KERNEL_map, "High Kernel Mapping" },
105 [MODULES_VADDR_NR] = { MODULES_VADDR, "Modules" },
106 [MODULES_END_NR] = { MODULES_END, "End Modules" },
107 [FIXADDR_START_NR] = { FIXADDR_START, "Fixmap Area" },
108 [END_OF_SPACE_NR] = { -1, NULL }
109};
110
111#define INIT_PGD ((pgd_t *) &init_top_pgt)
112
113#else /* CONFIG_X86_64 */
114
115enum address_markers_idx {
116 USER_SPACE_NR = 0,
117 KERNEL_SPACE_NR,
118 VMALLOC_START_NR,
119 VMALLOC_END_NR,
120#ifdef CONFIG_HIGHMEM
121 PKMAP_BASE_NR,
122#endif
123#ifdef CONFIG_MODIFY_LDT_SYSCALL
124 LDT_NR,
125#endif
126 CPU_ENTRY_AREA_NR,
127 FIXADDR_START_NR,
128 END_OF_SPACE_NR,
129};
130
131static struct addr_marker address_markers[] = {
132 [USER_SPACE_NR] = { 0, "User Space" },
133 [KERNEL_SPACE_NR] = { PAGE_OFFSET, "Kernel Mapping" },
134 [VMALLOC_START_NR] = { 0UL, "vmalloc() Area" },
135 [VMALLOC_END_NR] = { 0UL, "vmalloc() End" },
136#ifdef CONFIG_HIGHMEM
137 [PKMAP_BASE_NR] = { 0UL, "Persistent kmap() Area" },
138#endif
139#ifdef CONFIG_MODIFY_LDT_SYSCALL
140 [LDT_NR] = { 0UL, "LDT remap" },
141#endif
142 [CPU_ENTRY_AREA_NR] = { 0UL, "CPU entry area" },
143 [FIXADDR_START_NR] = { 0UL, "Fixmap area" },
144 [END_OF_SPACE_NR] = { -1, NULL }
145};
146
147#define INIT_PGD (swapper_pg_dir)
148
149#endif /* !CONFIG_X86_64 */
150
151/* Multipliers for offsets within the PTEs */
152#define PTE_LEVEL_MULT (PAGE_SIZE)
153#define PMD_LEVEL_MULT (PTRS_PER_PTE * PTE_LEVEL_MULT)
154#define PUD_LEVEL_MULT (PTRS_PER_PMD * PMD_LEVEL_MULT)
155#define P4D_LEVEL_MULT (PTRS_PER_PUD * PUD_LEVEL_MULT)
156#define PGD_LEVEL_MULT (PTRS_PER_P4D * P4D_LEVEL_MULT)
157
158#define pt_dump_seq_printf(m, to_dmesg, fmt, args...) \
159({ \
160 if (to_dmesg) \
161 printk(KERN_INFO fmt, ##args); \
162 else \
163 if (m) \
164 seq_printf(m, fmt, ##args); \
165})
166
167#define pt_dump_cont_printf(m, to_dmesg, fmt, args...) \
168({ \
169 if (to_dmesg) \
170 printk(KERN_CONT fmt, ##args); \
171 else \
172 if (m) \
173 seq_printf(m, fmt, ##args); \
174})
175
176/*
177 * Print a readable form of a pgprot_t to the seq_file
178 */
179static void printk_prot(struct seq_file *m, pgprotval_t pr, int level, bool dmsg)
180{
181 static const char * const level_name[] =
182 { "pgd", "p4d", "pud", "pmd", "pte" };
183
184 if (!(pr & _PAGE_PRESENT)) {
185 /* Not present */
186 pt_dump_cont_printf(m, dmsg, " ");
187 } else {
188 if (pr & _PAGE_USER)
189 pt_dump_cont_printf(m, dmsg, "USR ");
190 else
191 pt_dump_cont_printf(m, dmsg, " ");
192 if (pr & _PAGE_RW)
193 pt_dump_cont_printf(m, dmsg, "RW ");
194 else
195 pt_dump_cont_printf(m, dmsg, "ro ");
196 if (pr & _PAGE_PWT)
197 pt_dump_cont_printf(m, dmsg, "PWT ");
198 else
199 pt_dump_cont_printf(m, dmsg, " ");
200 if (pr & _PAGE_PCD)
201 pt_dump_cont_printf(m, dmsg, "PCD ");
202 else
203 pt_dump_cont_printf(m, dmsg, " ");
204
205 /* Bit 7 has a different meaning on level 3 vs 4 */
206 if (level <= 3 && pr & _PAGE_PSE)
207 pt_dump_cont_printf(m, dmsg, "PSE ");
208 else
209 pt_dump_cont_printf(m, dmsg, " ");
210 if ((level == 4 && pr & _PAGE_PAT) ||
211 ((level == 3 || level == 2) && pr & _PAGE_PAT_LARGE))
212 pt_dump_cont_printf(m, dmsg, "PAT ");
213 else
214 pt_dump_cont_printf(m, dmsg, " ");
215 if (pr & _PAGE_GLOBAL)
216 pt_dump_cont_printf(m, dmsg, "GLB ");
217 else
218 pt_dump_cont_printf(m, dmsg, " ");
219 if (pr & _PAGE_NX)
220 pt_dump_cont_printf(m, dmsg, "NX ");
221 else
222 pt_dump_cont_printf(m, dmsg, "x ");
223 }
224 pt_dump_cont_printf(m, dmsg, "%s\n", level_name[level]);
225}
226
227static void note_wx(struct pg_state *st, unsigned long addr)
228{
229 unsigned long npages;
230
231 npages = (addr - st->start_address) / PAGE_SIZE;
232
233#ifdef CONFIG_PCI_BIOS
234 /*
235 * If PCI BIOS is enabled, the PCI BIOS area is forced to WX.
236 * Inform about it, but avoid the warning.
237 */
238 if (pcibios_enabled && st->start_address >= PAGE_OFFSET + BIOS_BEGIN &&
239 addr <= PAGE_OFFSET + BIOS_END) {
240 pr_warn_once("x86/mm: PCI BIOS W+X mapping %lu pages\n", npages);
241 return;
242 }
243#endif
244 /* Account the WX pages */
245 st->wx_pages += npages;
246 WARN_ONCE(__supported_pte_mask & _PAGE_NX,
247 "x86/mm: Found insecure W+X mapping at address %pS\n",
248 (void *)st->start_address);
249}
250
251static void effective_prot(struct ptdump_state *pt_st, int level, u64 val)
252{
253 struct pg_state *st = container_of(pt_st, struct pg_state, ptdump);
254 pgprotval_t prot = val & PTE_FLAGS_MASK;
255 pgprotval_t effective;
256
257 if (level > 0) {
258 pgprotval_t higher_prot = st->prot_levels[level - 1];
259
260 effective = (higher_prot & prot & (_PAGE_USER | _PAGE_RW)) |
261 ((higher_prot | prot) & _PAGE_NX);
262 } else {
263 effective = prot;
264 }
265
266 st->prot_levels[level] = effective;
267}
268
269/*
270 * This function gets called on a break in a continuous series
271 * of PTE entries; the next one is different so we need to
272 * print what we collected so far.
273 */
274static void note_page(struct ptdump_state *pt_st, unsigned long addr, int level,
275 u64 val)
276{
277 struct pg_state *st = container_of(pt_st, struct pg_state, ptdump);
278 pgprotval_t new_prot, new_eff;
279 pgprotval_t cur, eff;
280 static const char units[] = "BKMGTPE";
281 struct seq_file *m = st->seq;
282
283 new_prot = val & PTE_FLAGS_MASK;
284 if (!val)
285 new_eff = 0;
286 else
287 new_eff = st->prot_levels[level];
288
289 /*
290 * If we have a "break" in the series, we need to flush the state that
291 * we have now. "break" is either changing perms, levels or
292 * address space marker.
293 */
294 cur = st->current_prot;
295 eff = st->effective_prot;
296
297 if (st->level == -1) {
298 /* First entry */
299 st->current_prot = new_prot;
300 st->effective_prot = new_eff;
301 st->level = level;
302 st->marker = address_markers;
303 st->lines = 0;
304 pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
305 st->marker->name);
306 } else if (new_prot != cur || new_eff != eff || level != st->level ||
307 addr >= st->marker[1].start_address) {
308 const char *unit = units;
309 unsigned long delta;
310 int width = sizeof(unsigned long) * 2;
311
312 if (st->check_wx && (eff & _PAGE_RW) && !(eff & _PAGE_NX))
313 note_wx(st, addr);
314
315 /*
316 * Now print the actual finished series
317 */
318 if (!st->marker->max_lines ||
319 st->lines < st->marker->max_lines) {
320 pt_dump_seq_printf(m, st->to_dmesg,
321 "0x%0*lx-0x%0*lx ",
322 width, st->start_address,
323 width, addr);
324
325 delta = addr - st->start_address;
326 while (!(delta & 1023) && unit[1]) {
327 delta >>= 10;
328 unit++;
329 }
330 pt_dump_cont_printf(m, st->to_dmesg, "%9lu%c ",
331 delta, *unit);
332 printk_prot(m, st->current_prot, st->level,
333 st->to_dmesg);
334 }
335 st->lines++;
336
337 /*
338 * We print markers for special areas of address space,
339 * such as the start of vmalloc space etc.
340 * This helps in the interpretation.
341 */
342 if (addr >= st->marker[1].start_address) {
343 if (st->marker->max_lines &&
344 st->lines > st->marker->max_lines) {
345 unsigned long nskip =
346 st->lines - st->marker->max_lines;
347 pt_dump_seq_printf(m, st->to_dmesg,
348 "... %lu entr%s skipped ... \n",
349 nskip,
350 nskip == 1 ? "y" : "ies");
351 }
352 st->marker++;
353 st->lines = 0;
354 pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
355 st->marker->name);
356 }
357
358 st->start_address = addr;
359 st->current_prot = new_prot;
360 st->effective_prot = new_eff;
361 st->level = level;
362 }
363}
364
365static void ptdump_walk_pgd_level_core(struct seq_file *m,
366 struct mm_struct *mm, pgd_t *pgd,
367 bool checkwx, bool dmesg)
368{
369 const struct ptdump_range ptdump_ranges[] = {
370#ifdef CONFIG_X86_64
371 {0, PTRS_PER_PGD * PGD_LEVEL_MULT / 2},
372 {GUARD_HOLE_END_ADDR, ~0UL},
373#else
374 {0, ~0UL},
375#endif
376 {0, 0}
377};
378
379 struct pg_state st = {
380 .ptdump = {
381 .note_page = note_page,
382 .effective_prot = effective_prot,
383 .range = ptdump_ranges
384 },
385 .level = -1,
386 .to_dmesg = dmesg,
387 .check_wx = checkwx,
388 .seq = m
389 };
390
391 ptdump_walk_pgd(&st.ptdump, mm, pgd);
392
393 if (!checkwx)
394 return;
395 if (st.wx_pages)
396 pr_info("x86/mm: Checked W+X mappings: FAILED, %lu W+X pages found.\n",
397 st.wx_pages);
398 else
399 pr_info("x86/mm: Checked W+X mappings: passed, no W+X pages found.\n");
400}
401
402void ptdump_walk_pgd_level(struct seq_file *m, struct mm_struct *mm)
403{
404 ptdump_walk_pgd_level_core(m, mm, mm->pgd, false, true);
405}
406
407void ptdump_walk_pgd_level_debugfs(struct seq_file *m, struct mm_struct *mm,
408 bool user)
409{
410 pgd_t *pgd = mm->pgd;
411#ifdef CONFIG_PAGE_TABLE_ISOLATION
412 if (user && boot_cpu_has(X86_FEATURE_PTI))
413 pgd = kernel_to_user_pgdp(pgd);
414#endif
415 ptdump_walk_pgd_level_core(m, mm, pgd, false, false);
416}
417EXPORT_SYMBOL_GPL(ptdump_walk_pgd_level_debugfs);
418
419void ptdump_walk_user_pgd_level_checkwx(void)
420{
421#ifdef CONFIG_PAGE_TABLE_ISOLATION
422 pgd_t *pgd = INIT_PGD;
423
424 if (!(__supported_pte_mask & _PAGE_NX) ||
425 !boot_cpu_has(X86_FEATURE_PTI))
426 return;
427
428 pr_info("x86/mm: Checking user space page tables\n");
429 pgd = kernel_to_user_pgdp(pgd);
430 ptdump_walk_pgd_level_core(NULL, &init_mm, pgd, true, false);
431#endif
432}
433
434void ptdump_walk_pgd_level_checkwx(void)
435{
436 ptdump_walk_pgd_level_core(NULL, &init_mm, INIT_PGD, true, false);
437}
438
439static int __init pt_dump_init(void)
440{
441 /*
442 * Various markers are not compile-time constants, so assign them
443 * here.
444 */
445#ifdef CONFIG_X86_64
446 address_markers[LOW_KERNEL_NR].start_address = PAGE_OFFSET;
447 address_markers[VMALLOC_START_NR].start_address = VMALLOC_START;
448 address_markers[VMEMMAP_START_NR].start_address = VMEMMAP_START;
449#ifdef CONFIG_MODIFY_LDT_SYSCALL
450 address_markers[LDT_NR].start_address = LDT_BASE_ADDR;
451#endif
452#ifdef CONFIG_KASAN
453 address_markers[KASAN_SHADOW_START_NR].start_address = KASAN_SHADOW_START;
454 address_markers[KASAN_SHADOW_END_NR].start_address = KASAN_SHADOW_END;
455#endif
456#endif
457#ifdef CONFIG_X86_32
458 address_markers[VMALLOC_START_NR].start_address = VMALLOC_START;
459 address_markers[VMALLOC_END_NR].start_address = VMALLOC_END;
460# ifdef CONFIG_HIGHMEM
461 address_markers[PKMAP_BASE_NR].start_address = PKMAP_BASE;
462# endif
463 address_markers[FIXADDR_START_NR].start_address = FIXADDR_START;
464 address_markers[CPU_ENTRY_AREA_NR].start_address = CPU_ENTRY_AREA_BASE;
465# ifdef CONFIG_MODIFY_LDT_SYSCALL
466 address_markers[LDT_NR].start_address = LDT_BASE_ADDR;
467# endif
468#endif
469 return 0;
470}
471__initcall(pt_dump_init);